Integrated Forecast of Financial Time Series Data Based on Deep Learning Model

Abstract

With the vigorous development of the market economy, effective prediction and analysis of the financial market has become an important part of people's daily life. FTS has massive historical data, and raw data preprocessing is very effective for improving system performance and model generalization ability. There are also many FTS forecasting techniques, including ARIMA, ARCH method of linear regression model, statistical regression and so on. However, most of the current forecasting models ignore the important step of extracting the intrinsic characteristics of time series, and directly perform regression forecasting, resulting in a significant reduction in forecasting accuracy and unsatisfactory results. The BP NN based on the deep learning model has excellent nonlinear characteristics, as well as the characteristics of rapid convergence and avoidance of local optimum. Therefore, this paper establishes the BPNN model as a prediction model for regression prediction. Since the stock index series is a typical representative of FTS, this paper selects the historical data of a stock composite index in the past five years as the training data, and uses the evaluation indicators such as MSE, MAPE and MAE to compare the PEs of the ARIMA model and the BPNN model. The results show that the BPNN model in this paper is more suitable for integrated forecasting of financial time series (FTS) data.